The role of interoception in the overlap between eating disorders and autism: Methodological considerations.

Significant comorbidity has been demonstrated between feeding and eating disorders and autism. Atypical interoception (perception of bodily signals) may, at least in part, be responsible for this association, as it has been implicated in the aetiology of both conditions. However, significant methodological limitations are impeding progress in this area. This paper provides a brief overview of how interoception has been linked to autism and feeding and eating disorders in both adolescent and adult populations before identifying several issues with current measures of interoception. We suggest that methodological issues may be contributing to the inconsistency in the empirical literature, and provide suggestions for future research.

six‐alternative forced‐choice

anorexia nervosa

bulimia nervosa

eating disorder

heartbeat counting task

heartbeat discrimination task

Interoceptive Accuracy Scale

Interoceptive Attention Scale

method of constant stimulation

other specified feeding and eating disorders

phase adjustment task

INTRODUCTION

Feeding and eating disorders (henceforth EDs) are characterised by disturbed eating and eating‐related behaviours that significantly impair one's physical and mental health (American Psychiatric Association, 2013). EDs have the highest mortality rates of all mental health disorders and exhibit a concerning resistance to treatment (Treasure, 2019; van Eeden et al., 2021), highlighting a pressing need for a greater understanding of their aetiological mechanisms.

Autism spectrum disorder (henceforth ‘autism’) is defined by social difficulties and restricted and repetitive interests and behaviours (American Psychiatric Association, 2013). We shall refer to those with autism as ‘autistic’, as preferred by the autistic community (Kenny et al., 2016). There is a considerable degree of co‐occurrence between autism and EDs—a higher proportion of both adult and adolescent ED patients score above clinically significant thresholds on autism measures (Carpita et al., 2020; Huke et al., 2013), and adults and adolescents with EDs score higher on continuous measures of autistic traits when compared to the non‐ED population (Baron‐Cohen et al., 2013; Dell’Osso et al., 2018; Karjalainen et al., 2019; Westwood et al., 2016). Conversely, autistic adults have higher rates of EDs than non‐autistic adults (Karjalainen et al., 2016). Furthermore, in adults, autistic traits correlate with ED symptoms in both ED patients and the general population (Dell’Osso et al., 2018). Elevated autistic traits have been observed in recovered adults and adolescents with EDs (Bentz et al., 2017; Karjalainen et al., 2019; Nazar et al., 2018), and ED patients report noticing autistic traits before the emergence of their ED (Mandy & Tchanturia, 2015; Westwood et al., 2018), suggesting that the association between EDs, especially anorexia nervosa (AN), and autistic traits is not solely due to starvation (Hiller & Pellicano, 2013). Potential reasons for the co‐occurrence of ED and autism have been discussed previously (Brede et al., 2020; Carpita et al., 2020), but here we focus on the hypothesised role of interoception in the aetiology of both conditions and their overlap, and outline methodological issues to be addressed when investigating this—particularly the measurement of interoception. Given the hypothesised clinical potential for interoceptive training, assessing the evidential basis of interoception's role in autism and ED is crucial. The majority of research on interoception and autism in EDs concerns AN, bulimia nervosa (BN) and other specified feeding and eating disorders (OSFED), and thus we focus mainly on these conditions.

INTEROCEPTION

Interoception is broadly considered to be a multidimensional construct encompassing the perception of the body's internal state (e.g., perceiving how fast one's heart is beating, or one's level of satiety; Craig, 2002; Ceunen et al., 2016; Murphy et al., 2017; Brewer et al., 2021). Interoception can be conceptualised within the model suggested by Murphy et al. (2019).

This model dissociates between the accuracy of an individual's perception of their internal states, and how much attention they pay to internal signals, both of which may be clinically relevant (Murphy et al., 2019). Furthermore, one can distinguish between an individual's objective degree of accuracy when perceiving, or their objective degree of attention to, internal signals, and their subjective estimates of, or beliefs about, these values (see Figure 1). Whether subjective beliefs relating to interoception are consistent with objective measures (e.g., if an individual believes they have accurate interoceptive perception when they do not, sometimes referred to as meta‐cognition) is thought to be of clinical relevance, specifically within the context of anxiety (Garfinkel et al., 2015), although this may apply to psychiatric conditions more broadly. The potential relevance to EDs can be illustrated with the example of intuitive eating. If an individual correctly believes they can accurately perceive signals of hunger and satiety, eating intuitively will allow them to meet their calorific needs. However, if an individual believes their perception is accurate when it is not, they may choose to eat intuitively rather than to rely on established meal times or calorific targets, which could result in them over or under‐eating. In addition to the above, one's evaluations of particular interoceptive signals may also be of clinical relevance (Herbert et al., 2020; Murphy et al., 2020; Suksasilp & Garfinkel, 2022). For example, one may interpret satiety signals as aversive, which may result in avoidance of food.

FIGURE 1

The dimensions upon which individual differences in interoception can be observed (from Murphy et al., 2019)

Interoception may vary across domains (where ‘domain’ refers to a specific interoceptive signal or group of signals; Murphy et al., 2017). Although some evidence suggests that there is a degree of coherence in accuracy of perception across domains (e.g., between cardiac and gastric interoception; Herbert et al., 2012; Whitehead & Drescher, 1980), other evidence suggests a lack of coherence (Ferentzi et al., 2018; Garfinkel et al., 2016; Pollatos et al., 2016). With respect to attention, the degree of coherence across domains is currently unclear.

HYPOTHESISED ROLE OF INTEROCEPTION IN AUTISM, EDS AND THEIR OVERLAP

Several theories suggest that atypical interoception has a causal role in the aetiology of autism (DuBois et al., 2016; Garfinkel et al., 2016; Quattrocki & Friston, 2014; but see Brewer et al., 2015), and EDs (Fassino et al., 2004; Herbert et al., 2020). However, these theories vary substantially in terms of the specific dimensions of interoception implicated, and the causal mechanisms believed to underlie the association between symptoms and atypical interoception. With respect to the role interoception may play in the overlap between EDs and autism, it may be that atypical interoception is associated with autistic symptoms or traits, which in turn predispose an individual to EDs. For example, autistic individuals may have high levels of interoceptive attention due to sensory sensitivity, to a degree which is unpleasant and overwhelming. This may result in them either avoiding situations like eating, where internal sensations such as satiety are heightened (Brede et al., 2020), or deliberately suppressing internal signals over time (Brewer et al., 2021), which may increase vulnerability to EDs. Alternatively, reduced objective interoceptive accuracy in autism could lead to difficulties in regulating food intake, or difficulties with emotion regulation, which could result in the use of abnormal eating behaviour as a coping strategy (Brede et al., 2020; Herbert et al., 2020). Further work is required to establish a more formal theory of the role of interoception in the overlap between EDs and autism. If it is the case that atypical interoception is responsible for the symptoms of either condition, or the association between them, then given that initial evidence suggests interoception may be improved with intervention (Quadt et al., 2021), interoception may represent a promising clinical target (see Carrard et al., 2011; Khoury et al., 2018). This may be especially relevant to individuals with both autism and EDs, who often report that existing models of ED therapy do not work for them, and demonstrate worse outcomes after intervention than non‐autistic patients (Brede et al., 2020; Kinnaird et al., 2019; Tchanturia, 2021).

Empirical research: Interoception in autism

Empirical research exploring interoception in autism has produced conflicting results. Three papers have reported higher self‐reported attention to interoceptive signals in autistic adults (Fiene & Brownlow, 2015; Garfinkel et al., 2016; Zdankiewicz‐Ścigała et al., 2021), but not children (Palser et al., 2018). Less accurate perception of cardiac signals in autistic adults (Garfinkel et al., 2016) and children (Nicholson et al., 2019; Palser et al., 2018) has also been reported, but other studies have found typical perception in autistic adults (Nicholson et al., 2018; Schauder et al., 2015). In terms of objective respiratory accuracy, one study found typical performance in autistic adults, although the findings are of questionable validity due to the lack of a control task (Nicholson et al., 2019).

Empirical research: Interoception in eating disorders

Results of studies of interoception in ED have been similarly mixed (Herbert, 2020; Jacquemot & Park, 2020; Klabunde et al., 2017; Martin et al., 2019). One meta‐analysis demonstrated an effect whereby both adolescent and adult ED patients reported worse perception of gastric and hunger‐related signals in everyday life, with a greater deficit in younger participants (Jenkinson et al., 2018). Similar results have been observed when considering self‐reported attention across interoceptive domains in both adolescents and adults (Brown et al., 2017). Pollatos et al. (2008) reported impaired perception of cardiac signals in AN patients; however, several studies have failed to replicate this result (Eshkevari et al., 2014; Kinnaird et al., 2020; Lutz et al., 2019). Kinnaird et al. (2020) reported that AN patients believed their accuracy of perception to be impaired, and these beliefs correlated with illness severity. In contrast to a control group, beliefs were not correlated with accuracy of perception in the AN group. Other studies have failed to find a reduction in perceived accuracy in AN patients (Lutz et al., 2019). There is some evidence, however, that adults with EDs may evaluate interoceptive signals more negatively, as during breath‐holding exercises participants with ED reported more stress and suffocation fear than healthy controls (Lapidus et al., 2020).

One limitation of the majority of work on interoception in ED is that interoception is only measured at one time point, despite evidence of state like variations in interoception in the general population (Wittkamp et al., 2018), as well as evidence that within ED patients specifically, the correspondence between perceived cardiac and respiratory intensity and heart rate fluctuates according to factors such as proximity to meal times (Khalsa et al., 2015). This also limits conclusions that can be drawn about the direction of influence, although preliminary longitudinal work has found self‐reported interoceptive attention to predict later ED (Bizeul et al., 2001; Clausen et al., 2011). Even disregarding this limitation, it is clear that research exploring the role of interoception in both autism and ED has produced extremely heterogeneous findings. We hypothesise that this heterogeneity is a result of the use of unreliable tests with limited validity. We suggest that before we can begin to look to develop interoceptive interventions, research testing the role of interoception in autism, EDs, and their association must overcome several methodological issues.

METHODOLOGICAL ISSUES

Measurement of interoception

Self‐report measures

Low agreement between self‐report measures make their interpretation difficult (Desmedt et al., 2021). Within ED research a substantial number of studies have assessed interoception using the Interoceptive Accuracy Subscale of the Eating Disorder Inventory (EDI, Garner et al., 1983), where respondents report their ratings of confidence in identifying emotions and sensations of hunger. It has been criticised as difficulties in determining one's emotional state pertain more to alexithymia (a sub‐clinical condition characterised by difficulty identifying or describing one's own emotions; Nemiah et al., 1976) than interoception, and the sole focus on hunger perception prohibits a generalised measure of confidence in interoceptive perception (Kinnaird et al., 2020). Similarly, the Multidimensional Assessment of Interoceptive Awareness, which has been used test interoceptive ‘awareness’ in both ED and autistic populations (Brown et al., 2017; Quadt et al., 2021), includes items that pertain to alexithymia and anxiety sensitivity rather than pure interoception (Mehling et al., 2012). A commonly used questionnaire in both ED and autism research is the Body Perception Questionnaire (Porges, 1993). This has been argued to conflate beliefs about interoceptive accuracy and attention (Murphy et al., 2019), with participants showing a large degree of heterogeneity in how they interpret the items (Gabriele et al., 2022). In response, the Interoceptive Accuracy Scale (IAS) and the Interoceptive Attention Scale (IATS) have since been developed as pure measures of beliefs about interoceptive accuracy and attention, respectively (Gabriele et al., 2022; Murphy et al., 2020). The IAS and IATS represent the most specific measures of beliefs about one's own interoceptive profile currently available. However, both measures are still undergoing refinement and as such have not yet been used or validated in either autistic or ED populations.

Objective measures

Cardiac

Problematically, the majority of experimental work on cardiac interoception in EDs and autism has used the heartbeat counting task (HCT, Schandry, 1981), or its variant the heartbeat tapping task (de Pascalis et al., 1984). Critiques of this approach have been well documented (Desmedt et al., 2018; Murphy et al., 2018; Windman et al., 1999), and as such a comprehensive evaluation is not necessary here. The most salient issue is that participants could achieve perfect performance on the task without being able to perceive their heartbeats at all if they have prior knowledge about what their heart rate should be (Ring & Brener, 1996; Ring et al., 2015).

An alternative to the HCT is the heartbeat discrimination task (HDT; Whitehead et al., 1977), in which participants judge which of two sets of auditory or visual tones is synchronous with their heartbeats. This method has been employed in both autistic and ED samples (Eshkevari et al., 2014; Nicholson et al., 2018). However, the test relies on the—false (Brener et al., 1993)—assumption that there are no individual differences in the delay following cardiac contraction with which individuals perceive their heartbeat. There are variations of the HDT that aim to address this issue by presenting tones at multiple different delays after cardiac contraction and assessing the consistency of synchronicity judgements (e.g. the six‐alternative forced‐choice [6AFC] and method of constant stimulation [MCS] methods; Brener et al., 1993; Clemens, 1984; Yates et al., 1985). These HDT variations—as long as they are coupled with appropriate control tasks assessing the participant's ability to judge the synchronicity of two stimuli (Knapp et al., 1997)—provide adequate tests of cardiac interoception but have not been used to test interoception in either ED or autism. This may be due to practical limitations: both measures are time‐consuming, and performance is thus constrained by participants' ability to focus for an extended period, which may be impaired in EDs such as AN and BN (Bosanac et al., 2007).

The phase adjustment task (PAT; Plans et al., 2020) is a new test of cardiac interoception that both overcomes the use of pre‐defined delays to test perception of heartbeats and is quicker than the 6AFC and MCS, therefore representing an advance on the HDT variants described above. The task presents participants with auditory tones that are at the same frequency as their heartbeats, but out of phase with those heartbeats. Participants use a dial to change the phase relationship between the tone and their heartbeat until they perceive them as synchronous. Interoceptive accuracy is determined by the consistency of their responses. The PAT is amenable for smartphone administration, increasing its acceptability for use in clinical populations (especially those that may not wish to travel to novel laboratory environments). The design of the task also prohibits false positives (identifying participants as interoceptive who are not), and the use of a closely matched screening task prevents false negatives (identifying participants as not interoceptive when they in fact are). As of yet, the PAT not been used in either ED or autistic participants.

Finally, the heartrate discrimination task (Legrand et al., 2021) is a new measure in the cardiac domain that does not attempt to measure the accuracy of perception of cardiac signals, or the degree of attention paid to them. Instead, the task aims to measure the accuracy and precision of beliefs about heart rate. Whether individual differences in these parameters are of clinical relevance is yet to be established, but in its current instantiation, the heartrate discrimination task requires a well‐matched control task to address questions relating to clinical relevance.

Respiratory

Respiratory interoception has been measured by both detection and discrimination tasks (Harrison et al., 2021; Webster & Colrain, 2000; Zhao et al., 2002), in which participants are required to detect whether respiration is restricted using filters, or to judge which of two filters provides more respiratory resistance, respectively. In both of these tasks, findings may be confounded by the fact that longer respiratory restrictions have an intrinsic negative value. More recently, a task has been developed that overcomes this, by measuring individuals' ability to detect small differences in lengths of short respiratory occlusions (van den Houte et al., 2021), also reducing the time taken to run the experiment. This measure has not yet been used in autistic or ED populations. Pragmatically, administration of any respiratory interoception task requires the use of specialist equipment, such as clinical breathing filters, which makes it difficult to administer remotely or at scale.

Gastric

Of high relevance to the field of EDs is the perception of gastric signals of satiety and fullness. Khalsa and colleagues provide a comprehensive overview of research of this nature in ED patients (Khalsa et al., 2022). Given the relevance of these signals, it is unfortunate that the most valid methods to assess their perception are invasive in nature, involving forced gastric distension, and are cumbersome and unpleasant to administer (Whitehead & Drescher, 1980). As an alternative, water‐load tests are frequently performed in healthy populations (van Dyck et al., 2016), in which the amount of water required for participants to report satiety and a full stomach is measured. Water‐load tests do not provide a measure of the accuracy of perception of these signals, however, and may be considered a measure of beliefs about interoceptive signals and their relevance (in this case, the degree of water required to feel satiated), rather than the accuracy of interoceptive perception or the degree of attention paid to interoceptive signals. To our knowledge, none of these methods have been used in studies with autistic participants. With all of these methods, there are ethical issues associated with their use in both autistic participants and ED patients, where the induced feelings of fullness may be especially distressing.

Interoceptive perception or the perception of internal states?

For both tests of cardiac and respiratory accuracy, it is unknown whether participants' performances depend on the accurate detection of purely interoceptive signals, or other, non‐interoceptive cues. Presumably, in respiratory tasks, participants detect a discrepancy between the expected and experienced degree of airflow for a given degree of respiratory (muscular) effort. The signal used to detect this discrepancy may result from any combination of stretch receptors in the skin of the mouth or chest, intra‐oral or extra‐oral pressure, changes in temperature of the airway, or numerous other signals. Similarly, cardiac perception tasks may rely on either interoceptive signals from baroreceptors, or non‐interoceptive signals such as chest wall vibrations (Khalsa et al., 2009; Murphy et al., 2018). This is important if one considers that it is only perception of interoceptive signals that is of clinical relevance, but it is equally likely (Brewer et al., 2021; Ceunen et al., 2016) that what matters is perception of the signal itself, and not which receptors are involved. Addressing this question is a priority for future research, especially if the perception of interoceptive signals via exteroceptive means can ameliorate any symptoms associated with either autism or ED.

CONCLUSION

Given the reported associations between various EDs and autism, and the fact that interoception has been implicated in both conditions, it is feasible that interoception may contribute to the observed link between autism and EDs. However, current research does not provide compelling evidence for this, due to the limitations of the methodologies employed. The most pertinent issue is that current measures of interoception lack validity. Before research can be conducted into the potential for interoception based‐interventions in autistic ED patients, more research is required to test and develop promising novel measures of interoception across domains and dimensions.

CONFLICT OF INTEREST

There is no conflict of interest to declare.